Brain-computer interfaces (BCI) based on steady-state visual evoked potentials (SSVEP) enable a user to control an application by focusing his/her attention on visual stimuli blinking at specific frequencies. This technique of interaction can enable people suffering from severe motor disabilities to improve their quality of life through regaining a partial autonomy. According to literature, each usage session of a SSVEP-based BCI integrates a calibration phase aimed in particular at computing classifier's parameters. Our objective is to evaluate if the same parameters could be used during several sessions, in order to avoid performing systematically a calibration phase, which is very restrictive for the user. To do so, we analyze stability of classification results over time. On the other hand, the data acquired during our experiments were used to study the possible effects of human learning on interface performance and to confirm or not the state of the art knowledge on this subject. According to literature, SSVEP-based BCIs work well from the first use and their performances do not improve with subject's experience.